Hannah Bull

Dr. Hannah Bull


Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 March 2024 in Remote Sensing
Verified
Journal Article
Open access
PDF file
AKAlisa Kugusheva
Alisa Kugusheva
HBHannah Bull
Hannah Bull
EMEvangelos Moschos
Evangelos Moschos
AIArtemis Ioannou
Artemis Ioannou
BVBriac Le Vu
Briac Le Vu
ASAlexandre Stegner
Alexandre Stegner
Real-time reconstruction of ocean surface currents is a challenge due to the complex, non-linear dynamics of the ocean, the small number of in situ measurements, and the spatio-temporal heterogeneity of satellite altimetry observations. To address this challenge, we introduce HIRES-CURRENTS-Net, an operational real-time convolutional neural network (CNN) model for daily ocean current reconstruction. This study focuses on the Mediterranean Sea, a region where operational models have great difficulty predicting surface currents. Notably, our model showcases higher accuracy compared to commonly used alternative methods. HIRES-CURRENTS-Net integrates high-resolution measurements from the infrared or visible spectrum—high resolution Sea Surface Temperature (SST) or chlorophyll (CHL) images—in addition to the low-resolution Sea Surface Height (SSH) maps derived from satellite altimeters. In the first stage, we apply a transfer learning method which uses a high-resolution numerical model to pre-train our CNN model on simulated SSH and SST data with synthetic clouds. The observation of System Simulation Experiments (OSSEs) offers us a sufficient training dataset with reference surface currents at very high resolution, and a model trained on this data can then be applied to real data. In the second stage, to enhance the real-time operational performance of our model over previous methods, we fine-tune the CNN model on real satellite data using a novel pseudo-labeling strategy. We validate HIRES-CURRENTS-Net on real data from drifters and demonstrate that our data-driven approach proves effective for real-time sea surface current reconstruction with potential operational applications such as ship routing.

26 March 2025
Verified
Conference Paper
Open access
ILInès Larroche
Inès Larroche
APAmélie Pesnec
Amélie Pesnec
PGPierre Garcia
Pierre Garcia
TAThéo Archambault
Théo Archambault
HBHannah Bull
Hannah Bull
We present a machine learning approach that delivers 7-day operational forecasts of ocean surface currents. Our neural network is trained exclusively on real satellite data and in-situ observations, eliminating the need for regional numerical models or observation system simulation experiments to learn ocean dynamics. We validate the model’s performance with in-situ data from drifters and demonstrate improved accuracy compared to commonly-used forecasting methods. Moreover, we validate our method using ADCP instruments on board ships, including data which will be collected on board the Statsraad Lehmkuhl during the ESA Advanced Ocean Training Course 2025. Providing accurate nowcasts and forecasts of ocean surface currents in real time is challenging due to the indirect and often incomplete nature of satellite remote sensing data. Our model is a multi-stage, multi-arm network specifically designed for high-resolution forecasts of ocean surface currents from various sparse or noisy data sources. Signatures of ocean surface currents are visible in high-resolution images of sea surface temperature (SST) and chlorophyll, and we thus use previous days of SST and chlorophyll as inputs to our model. Satellite altimetry from sparse Nadir altimeters and, more recently, from the dense, high-resolution SWOT altimeter also provide partial measures of ocean surface currents. We use past days of satellite altimetry as inputs to our model and future days of satellite altimetry as targets to train our model. The architecture is a multi-arm encoder-decoder neural network, trained in three stages. In the first stage, we learn to predict sea surface height and geostrophic ocean surface currents from decades of sparse Nadir altimetry data. In a second stage, we improve these predictions by training the model on one year of high-resolution satellite altimetry from the recent SWOT satellite, facilitating the accurate prediction of small-scale eddies and other features that are often missed by Nadir altimeters. Finally, in a third stage, we finetune our model using decades of sparse observations from drifters, which directly measure ocean surface currents. A positional encoding module enables the model to integrate temporal and geographic information.Our model has shown improved nowcasting and forecasting accuracy compared to other leading methods, particularly in regions with high kinetic energy and rapidly changing sea states. High-resolution prediction of ocean surface currents has many potential applications, including optimising ship routes, modelling climate dynamics and tracking of pollutants. Furthermore, using our ocean current model in combination with sea state observations provided by wind and wave sensors on board ships, we can capture critical wave-current interactions and predict extreme wave behaviour in dynamic ocean environments.